Robustness Checks

126

The interaction term between low level innovation projects and knowledge heterogeneity is introduced in model 3. The model tests hypothesis 2 stating that the relationship between foreign linkages and association with successful innovation project depends on the level of innovation. However, the model also re-tests the positive association proposed in hypothesis 1 by taking into account the interaction between innovation level and innovative performance.

Model 3 indicate that knowledge heterogeneity increase the likelihood that an individual is associated with a successful innovation project. But the interaction effect of knowledge heterogeneity and low-level innovation projects aimed at variety is significantly negative. A Wald chi2 test confirms that the magnitude of the effects differs and shows a significantly negative net effect of knowledge heterogeneity for participation in projects aimed at variety.

Participation in productions subsidized through the consultancy scheme, and productions distributed by Scandinavian/North European distributors are positively associated with the probability of a nomination. One of the matching variables shows significant within group variation: the average performance on foreign markets is significantly positively associated with the probability of being nominated.

127

Sweden, and Finland), Continental Europe (France and Germany), the US, Anglo-Saxon countries outside of the US (UK and New Zealand), and a residual category of other, less frequently occurring, countries. The results support the main results.

Furthermore, proximity and distance are not binary distinctions but rather a continuum. Some of the foreign contexts analyzed are more distant than others. The neighboring Scandinavian countries of Norway and Sweden, in particular, provide an accessible foreign context due to little geographical distance, low language barriers, and collaboration tradition. Geographical proximity increases the probability of tie formation (Sorenson and Stuart, 2008), and in our population a substantial part of the foreign linkages are with Norway and Sweden. 674 observations have foreign linkages with Norway or Sweden, while 382 have linkages with the US, 438 with other Anglo-Saxon countries (primarily the UK), and 555 have links with continental Europe.

There is a clear bias in favor of Scandinavia. These differences in geographical and social proximity should lead to differences in degree of foreignness of perspectives acquired through collaboration in these foreign environments (Boschma, 2005; Hong and Page, 2001; Page, 2007). Consequently, knowledge heterogeneity acquired through linkages to Scandinavian countries should have a lower estimated effect on the probability for participation in a successful innovation project. Models 4 to 7 investigate and find support for this claim.

The positive estimates on knowledge heterogeneity and association with a successful innovation project still holds. However, only in the case of non-Scandinavian foreign linkages do we find a significant effect.

A final issue is related to the role of selection by other agents in the process of foreign linkage creation. To form foreign linkages and acquire heterogeneous knowledge, individuals need to be invited (and must accept the invitation) to join a foreign project. Within the European film industry, the idea and vision for a new film project are primarily formed by the director who develops them with the help of screenwriters and producers (Delmestri, 2005; Strandvad, 2008). Subsequently, the other potential participants in the production are

128

invited, and casting and selection take place. The findings might be the result of foreign agent selection bias in the casting and selection process. Therefore the role of other agent’s selection is controlled for by estimating equivalent models for screenwriters and directors. The work of producers is context dependent because sets, non-creative collaboration partners, and local subsidy schemes comprise a substantial part of the context. Producers are therefore not included in this robustness check. Screenwriters and Directors enter the production process in the first phase and participate in defining the project.

They subsequently select and invite the remaining cast and crew. Analyses of these 458 screenwriters and directors confirm the results, which means our findings are not caused by bias in the process of selecting talent.

The intuition from linear models cannot be applied to nonlinear models such as the logit model (Ai and Norton, 2003; Wiersema and Bowen, 2009).

Therefore the analysis needs to extend the classic regression tables. Due to the logit function’s structure, marginal effects vary with the probability

distribution, and the optimal way to provide an overview of an estimated effect is thus a graphic representation. The main and moderating effects of models 3, 5, and 7 are displayed in table 4.3.

The main effect is positive for all observations in all three models. Though the association between knowledge heterogeneity and affiliation with a successful innovation projects is significantly positive in the estimated models, the effect is not significant for all the individual observations. The moderating effects are negative for all observations in all three models, but only significantly negative for the upper end of the probability distribution.

129

130

Marginal effects are calculated for meaningful values of all the variables based on the estimates in model 3. The most distinctive change in the marginal effect of knowledge heterogeneity is for differences in production budgets and previously generated revenue. The marginal effect of knowledge heterogeneity is a 27.6 percentage point increase in the probability of being associated with a successful innovation project for participants in a low budget film, and this decreases to a 24.4 percentage point increase for participants in a medium budget film, 16.4 percentage point increase for participants in a high budget film, and only a 7 percentage point increase in probability for participants in the few productions with extremely high budgets.

For participants in productions aimed at innovation through creation of variety rather than novelty, knowledge heterogeneity decreases the probability of association in a successful innovation by 78.2 percentage point for low budget productions, 78 percentage point for medium budget productions, 73.5 percentage point for high budget productions, and 54.6 percentage point for extremely high budget productions. Thus the magnitude of the marginal effects of the interaction term is affected less by budget size. The marginal effects in models 5 and 7 show similar trends. Marginal effects are larger for the early years of the analyzed period. The magnitude of the marginal effects decreases with increasing industry internationalization over time. The magnitude of the marginal effects also decreases when individual level average revenue (domestic and foreign) increases.